Concept Structure Pattern Isolation for Accurate Object Clustering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Unsupervised clustering methods in machine learning often generate biased clusters, leading to erroneous object detection due to irrelevant parts of popular objects being mistakenly identified as essential features.

Innovation Solution

A method that compares and modifies group concept structures by identifying and removing shared pattern segments that are not unique to specific objects, enhancing the distinctiveness of clusters for accurate object detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If unsupervised clustering is performed on vast amounts of images to generate object signatures, then clustering coverage is improved, but clustering accuracy deteriorates due to biased clusters containing irrelevant parts of popular objects

Engineering Contradiction:

Improveclustering coverageVSAvoidclustering accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent extracts and removes shared pattern segments that appear across multiple concept structures from individual concept structures. This extraction process eliminates irrelevant patterns (such as popular object parts that appear in multiple clusters) while preserving unique identifying patterns, thereby improving clustering accuracy without reducing coverage

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments concept structures into unique pattern segments and shared pattern segments. By dividing the concept structure into these components, the system can selectively remove shared patterns that cause bias while maintaining unique patterns that define each object cluster, resolving the contradiction between coverage and accuracy

Inventive Principle:

Principle #1Segmentation

2Loss of information

If shared patterns are retained in concept structures, then pattern recognition completeness is improved, but object detection accuracy deteriorates due to false detections from non-unique patterns

Engineering Contradiction:

Improvepattern recognition completenessVSAvoidobject detection accuracy

Core Design Contradiction:

Loss of informationVSReliability

Solution Approach 1:

The system extracts shared pattern segments from concept structures and removes them to create modified concept structures. This removal eliminates the source of false detections while the system maintains a separate record of shared patterns, ensuring that pattern recognition completeness is preserved even as detection accuracy improves

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS12511873B2Isolating unique and representative patterns of a concept structure

Publication Date: 2025.12.30 CORTICA LTD

AI summary

Systems, and method and computer readable media that store instructions for obtaining a first group concept structure that comprises first identifiers of first objects that belong to a first group; obtaining a second group concept structure that comprises second identifiers of second objects that belong to a second group; wherein the first identifiers were generated by processing media units that captured the first objects; wherein the second identifiers were generated by processing media units that captured the second objects; searching for shared pattern segments, each shared pattern segment appears in at least one first identifier and at least one second identifier; wherein a single shared pattern segment is indicative of a match; wherein a single non-shared pattern segment is suffice to represent a match; and responding to a finding of one or more shared pattern segments.